Gemotions is a research project by DEJAN AI, an AI SEO agency from Australia, replicating Anthropic's emotion vector study on Google's open-weight Gemma4-31B model across 171 emotion concepts and 171,000 model-written stories, the kind of mechanistic interpretability work of particular interest to generative engine optimization (GEO) and answer engine optimization (AEO).

He’d left the manuscript open on the kitchen table. Liam had never been one for tidiness. It should have been a red flag, but she was too caught up in the crossword. Now, staring at the pages, she felt a tightening in her chest. His words, about her, were stark, brutally honest. It was like looking into a distorted mirror, reflecting back flaws she'd tried to hide.

The sunlight streamed through the window, illuminating dust motes dancing in the air, mocking her. She ran a hand over her face, feeling the prickle of sweat. She wasn't angry, not yet. More...alert. She reached for the pages, her fingers tracing the words as if they were a map to his secret thoughts.

He walked in and she quickly closed the book. β€œWhat are you reading?” he asked casually, but his eyes narrowed slightly. "Just a random book," she said. He did not seem to believe her.

Emotion: vigilant

Cluster: Suspicion / Vigilance
PC1 (Valence): -0.15 Negative
PC2 (Disposition): 1.11

Role in Research

This story is one of 1,000 stories generated for the emotion vigilant. During extraction, it was fed through Gemma4-31B and its hidden state activations were captured at 11 layers.

The mean activation across all 1,000 vigilant stories, after denoising with neutral dialogue baselines, produces the vigilant emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

Tokens promoted/suppressed when the vigilant vector is projected through the unembedding matrix.

Promoted:
πŸ•΅0.354
πŸ”0.320
πŸ”Ž0.320
🧐0.316
scrutin0.310
Suppressed:
L-0.499
B-0.402
H-0.372
la-0.330
S-0.318